flowchart TD
step_Virtual_Screening["Virtual Screening"]
style step_Virtual_Screening stroke-width:2px
step_Prerequisites_and_preliminary_steps("Prerequisites and preliminary steps")
step_Choose_a_ligand_library_to_screen_against("Choose a ligand library to screen against")
step_Construct_the_filter_funnel{{"Construct the filter funnel"}}
step_Prepare_and_understand_available_structural_data("Prepare and understand available structural data")
step_Pose_independent_structure_based_filters{{"Pose-independent structure based filters"}}
step_SBph4("Pharmacophore screening")
step_Docking("Docking")
step_GlideWS("GlideWS")
step_Pose_dependent_structure_based_filters{{"Pose-dependent structure based filters"}}
step_MM_GBSA("MM-GBSA")
step_FEP("FEP+")
step_Choose_and_set_up_ligand_based_filters{{"Choose and set up ligand-based filters"}}
step_Shape_screening("Shape screening")
step_LBph4("Pharmacophore screening")
step_QSPR_modeling("QSPR modeling")
step_Screen_the_library_against_the_filter_funnel("Screen the library against the filter funnel")
step_Conclusion_and_next_steps("Conclusion and next steps")
step_Virtual_Screening --> step_Prerequisites_and_preliminary_steps
step_Prerequisites_and_preliminary_steps --> step_Prepare_and_understand_available_structural_data
step_Prepare_and_understand_available_structural_data --> step_Choose_a_ligand_library_to_screen_against
step_Choose_a_ligand_library_to_screen_against --> step_Construct_the_filter_funnel
step_Construct_the_filter_funnel --> step_Choose_and_set_up_ligand_based_filters
step_Shape_screening --> step_Pose_independent_structure_based_filters
step_LBph4 --> step_Pose_independent_structure_based_filters
step_QSPR_modeling --> step_Pose_independent_structure_based_filters
step_Pose_independent_structure_based_filters --> |" "| step_SBph4
step_Pose_independent_structure_based_filters --> |" "| step_Docking
step_Pose_independent_structure_based_filters --> |" "| step_GlideWS
step_SBph4 --> step_Pose_dependent_structure_based_filters
step_Docking --> step_Pose_dependent_structure_based_filters
step_GlideWS --> step_Pose_dependent_structure_based_filters
step_Pose_dependent_structure_based_filters --> |" "| step_MM_GBSA
step_Pose_dependent_structure_based_filters --> |" "| step_FEP
step_MM_GBSA --> step_Screen_the_library_against_the_filter_funnel
step_FEP --> step_Screen_the_library_against_the_filter_funnel
step_Choose_and_set_up_ligand_based_filters --> |" "| step_Shape_screening
step_Choose_and_set_up_ligand_based_filters --> |" "| step_LBph4
step_Choose_and_set_up_ligand_based_filters --> |" "| step_QSPR_modeling
step_Screen_the_library_against_the_filter_funnel --> step_Conclusion_and_next_steps
classDef path_title stroke-width:2px,fill:#12122c,stroke:#12122c
classDef decision_step stroke-width:2px,fill:#005aaa,stroke:#005aaa
classDef simple_step stroke-width:2px,fill:#12122c,stroke:#12122c
class step_Virtual_Screening path_title
class step_Construct_the_filter_funnel,step_Prerequisites_and_preliminary_steps,step_Choose_a_ligand_library_to_screen_against,step_Prepare_and_understand_available_structural_data,step_SBph4,step_Docking,step_MM_GBSA,step_GlideWS,step_FEP,step_Shape_screening,step_LBph4,step_QSPR_modeling,step_Screen_the_library_against_the_filter_funnel,step_Conclusion_and_next_steps simple_step
class step_Pose_dependent_structure_based_filters,step_Pose_independent_structure_based_filters,step_Choose_and_set_up_ligand_based_filters decision_step
Learning Path: Virtual Screening
Searching chemical spaces for chemical matter with certain properties can be challenging depending on the amount of available information. Building out a virtual screening funnel requires a good overview of the available data and methods and understanding of goals and systems. In theory, any computational method that can be run on a series of compounds and return some form of ranking can be used in virtual screening funnels. The limiting factor for which methods are useful is the computational cost of applying them to large ligand libraries. HTVS funnels frequently combine structure-based and ligand-based approaches to harness their respective strengths. In practice, a structure-based screening funnel usually includes rigid-receptor docking, and may also include more computationally costly methods such as Free Energy Perturbation (FEP) or GlideWS as additional filters. Ligand-based methods can either be used as pre-filters to structure-based steps (reducing large chemical spaces to manageable size) or in parallel to structure-based steps in a consensus approach (carrying forward screening hits from both methods).
Next step: Prerequisites and preliminary steps
Prerequisites and preliminary steps
Before designing a screening funnel, you should collect available structural and activity data about your target. Having at least some known active compounds is very helpful for validating the screening funnel. You will need to make decisions on and prepare both the screening funnel and the ligand library to screen against.Prepare and understand available structural data
If multiple experimental structures are available, you need to understand the differences between them and choose which one(s) to use. This choice requires understanding of the system and particularly of the flexibility of the binding site, as e.g. docking with Glide considers the receptor to be rigid. It is possible to screen against multiple structures, e.g. for a protein with multiple distinct conformations that affect the binding pocket. It is especially important to capture the biologically relevant protonation and tautomer states (and these should match the assay environment used to validate hits). If no experimental structure is available, it is possible to perform a virtual screen using a computationally predicted structure.
How to prepare ligand libraries and target structures for screening.
An overview of approaches for understanding target structures.
An overview of computational structure prediction methods.
Enabling protein structures from x-ray crystallography, cryo-EM, ML-methods, and homology modeling for structure-based computational workflows
Next step: Choose a ligand library to screen against
Choose a ligand library to screen against
Depending on the nature and stage of of your project, this can be an internal library of accessible chemistry, an ultra-large library from a vendor, a library of fragments, or the result of an enumeration to decorate a fixed core. The size and nature of the library places limitations on the screening methods used due to computational costs and inherent methodical limitations. Ligand files can be sourced from numerous places, such as vendors or databases, often in the form of 1D or 2D structures with unstandardized chemistry. For some of the methods listed below, ligand files must then be prepared, i.e. converted to 3D structures, with the chemistry properly standardized and extrapolated, ready for use in virtual screening. While the computational cost of preparing a ligand is not huge, it does make sense to run the ligand preparation right before the steps that need it (e.g. before docking).
How to prepare ligand libraries and target structures for screening.
How to generate libraries that explore a particular chemical space.
An introduction to considerations for library design.
Next step: Construct the filter funnel
Construct the filter funnel
You can now decide which ligand-based and structure-based methods to use in the funnel. This is not an either-or decision: Making use of both types of methods is helpful. Ligand-based filters can be highly computationally efficient and are predominantly used in the earlier stages of a screening funnel as they can handle very large libraries. Structure-based filters can be highly discriminating but due to the increased computational cost are frequently employed once the chemical space has been somewhat narrowed.
Next step: Choose and set up ligand based filters
Decide: Choose and set up ligand-based filters
Many ligand-based filters use heuristics like druglikeness or PAINS patterns to filter out compounds. These are highly specific to each project and we'll not cover them here. More generic ligand-based filters try to find similarities between compounds on different levels, e.g. shape or functional groups.
- go to Shape screening
- go to Pharmacophore screening
- go to QSPR modeling
Shape-based screening
There are many different screening methods based on structural similarities between known binders. These methods use a variety of representations, e.g. one-dimensional connectivity fingerprints, three-dimensional shape, or chemical functionality. A key aspect of their success is a good prediction of plausible ligand states and conformations.
Efficiently screening millions of compounds and analyzing the results
3D shape based screening
Next step: Pose-independent structure based filters
Pharmacophore screening
Ligand-based pharmacophore screening with Phase makes use of similar chemical functionality present in known actives combined with their locations in 3D space.
Next step: Pose-independent structure based filters
QSPR modeling
Machine learning based models of properties of interest (e.g. physicochemical or PK/PD properties) can be used to efficiently filter areas of chemical space which don't fit the target profile.
Next step: Pose-independent structure based filters
Decide: Choose and set up pose-independent structure based filters
Initial structure-based filters do not require a bound ligand pose. Validate each filter individually using known actives and decoys before assembling the funnel.
- go to Pharmacophore screening
- go to Docking
Pharmacophore screening
Phase pharmacophore screens can make use of structural information to build a pharmacophore hypothesis from ligand-receptor complexes.
Next step: Pose-dependent structure based filters
Docking
Glide docking is well-suited to scoring diverse ligand sets and generates a bound pose of the ligand, which is needed for the more accurate structure-based filters (e.g. FEP). A covalent docking protocol based on Glide is available as well. This filter is suitable for library sizes up to 105, but can be augmented with active learning to screen libraries with several million ligands. Additional tools like ConfGen or Prime are helpful for increasing conformational sampling for large or macrocyclical ligands.
Introduction to Glide
Increasing conformational sampling for tricky ligands.
Glide contains a variant of its scoring function optimized for RNA receptors.
Next step: Pose-dependent structure based filters
GlideWS
It is possible to dock ligands to a GlideWS model, which includes an ensemble of protein structures for assessment of small induced fit effects and protein reorganization. The scoring includes terms for water displacement and desolvation effects, molecular recognition (H-bonds, hydrophobic enclosure, etc), ligand strain, and salt bridges.
Next step: Pose-dependent structure based filters
Decide: Choose and set up pose-dependent structure based filters
These methods require a bound ligand pose, which can be obtained from Glide docking. They are also more computationally costly and are a bit more involved to set up and validate.
MM-GBSA
MM-GBSA can be used to approximate the free energy of binding between a protein and a ligand for a given binding pose using an implicit solvent model without employing molecular dynamics simulations.
Next step: Screen the library against the filter funnel
FEP+
FEP+ methods have significantly larger computational costs and require thorough validation on known actives, but can be very powerful filters for the last stage of a screening funnel. This filter is suitable for library sizes of a few hundred ligands but can be augmented by an active learning approach to handle several thousand compounds.
Next step: Screen the library against the filter funnel
Screen the library against the filter funnel
Before each filter, you need to perform the necessary preparation steps (e.g. calculating fingerprints, running LigPrep, ...). It can be interesting to build the funnel not just sequentially but use filters in parallel to leverage consensus. Increasingly more accurate filters are increasingly computationally costly -- balance this vs the cost of sourcing the compounds and wet-lab testing.
Next step: Conclusion and next steps
Conclusion and next steps
After the screening, you can analyze the results and validate your hits experimentally. You can then drill down on interesting areas of chemical space using the De Novo Design Workflow or probe SAR with FEP+ and start ideating and optimizing.